Data, Not Models, Is the Bottleneck for Physical AI

Language models learned from an internet-scale corpus of human knowledge. Physical AI has no equivalent dataset. The next generation of capable robots will depend on collecting vastly more high-quality, real-world interaction data.

Why We Instrument Humans Instead of Teleoperating Robots

If humans already perform physical tasks naturally, why force them to operate robots to generate training data? Instrumenting people allows data collection to happen at the speed, scale, and diversity of real human behavior.

The Fidelity Floor: What Manipulation Data Must Preserve

Not all motion data is useful for robot learning. Manipulation datasets must preserve the physical signals that define how an action actually happens—from finger articulation to precise 6-DoF trajectories and contact events.